Inspiration
Research is slow because papers, experiments, and tools remain disconnected. Lakuna turns research gaps into testable ideas.
What it does Autonomous agents search papers, analyze evidence, propose solutions, modify code, run experiments, and judge results.
How we built it Lakuna combines Python agents, a Bun/React dashboard, Nexla data ingestion, Zero-powered Exa search, and Akash deployment.
Challenges We handled unreliable LLM outputs, failed experiments, dependency issues, secure credentials, and agent fallback behavior.
What we learned Reliable autonomous research requires grounded evidence, specialized agents, strict evaluation, and safe recovery ~ not just generated ideas.
Built With
- akash-network
- anthropic
- arxiv
- bun
- docker
- exa
- grobid
- ml
- nexla
- openrouter
- python
- pytorch
- react
- tailwind
- typescript
- websockets
- zero.xyz
Log in or sign up for Devpost to join the conversation.